Automatic 3D Retinal Vessel Density Mapping with OCTA

Abstract

Purpose : OCTA quantitative analysis of vessel densities typically produce a few summary statistics in retinal layers or region of interests from enface images. Although previous methods have successfully detected significant retinal vessel density differences in comparing diabetic retinopathy (DR) patients with normal controls, there is no 3D technique in OCTA literature that can localize the regions with vascular alternations across populations. Furthermore, inaccuracy in OCT structure layer segmentation added another challenge to localized vessel density analysis, which is more difficult in presence of pathologies such as edema. The aim of this study is to provide a robust framework for cross-subject localized vessel density mapping in 3D-OCTA of DR patients even in the presence of edema or other pathologies that cause severer OCT layer deformations.

Methods : We propose an automated 3D framework for localized vessel density mapping in OCTA images. Initially, we calculate enhanced vessel map via curvelet denoising and OOF-based analysis. Then, a vessel density image (VDI) is constructed by diffusing the vessel mask derived from OOF-based analysis to the entire image volume. Ultimately, to bring the subjects in common space and also address the OCT layer segmentation inaccuracy challenge, we implement a non-linear 3D OCT image registration approach to provide localized comparison of retinal vasculature across subjects at the voxel-level. Fig.1(1-3) illustrates the three main steps of the proposed method.

Results : In our experimental results, we applied our method to a clinical dataset from patients with DR. Fig.1(4) demonstrates the result of comparing NCs (n=25) and NPDR (n=54) subjects in superficial and deep retinal layers at voxel-level. Among the NPDR subjects, 20% of the images have cystoid macular edema. Our results in the significant areas indicate that NPDR group has consistently lower vessel density than the NCs. The localized vascular analysis shows that deep vascular plexus appears to be affected more than the superficial plexus. In addition, the vascular density changes in SRL are more localized in the temporal area of the retina.

Conclusions : In this work we provide a robust and automatic 3D framework for localized analysis of retinal vasculature alternation in OCTA scans across population. The proposed method has been tested on a clinical dataset and show successful quantitative results for DR pathology in the presence of edema.

This is a 2020 Imaging in the Eye Conference abstract.